# 用 Tensorflow 可视化梯度下降

```python
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

LR=.1
REAL_PARAMS=[1.2,2.5]
INIT_PARAMS=[[5,4],
             [5,1],
             [2,4.5]][2]
x=np.linspace(-1,1,200,dtype=np.float32)

y_fun = lambda a, b: a * x + b
tf_y_fun = lambda a, b: a * x + b
noise=np.random.randn(200)/10
y=y_fun(*REAL_PARAMS)+noise


a,b=[tf.Variable(initial_value=p,dtype=tf.float32) for p in INIT_PARAMS]
pred=tf_y_fun(a,b)
mse=tf.reduce_mean(tf.square(y-pred))
train_op=tf.train.GradientDescentOptimizer(LR).minimize(mse)

a_list,b_list,cost_list=[],[],[]
with tf.Session() as sess:
    tf.global_variables_initializer().run()
    for i in range(400):
        a_,b_,mse_=sess.run([a,b,mse])
        a_list.append(a_)
        b_list.append(b_)
        cost_list.append(mse_)
        result,_=sess.run([pred,train_op])

print('a=',a_,'b=',b_)
plt.figure(1)
plt.scatter(x,y)
plt.plot(x,result,color='r',linewidth=5)

fig=plt.figure(2)
ax=Axes3D(fig)
a3D,b3D=np.meshgrid(np.linspace(-2,7,30),np.linspace(-2,7,30))
cost3D=np.array([np.mean(np.square(y_fun(a_,b_)-y))for a_,b_ in zip(a3D.flatten(),b3D.flatten())]).reshape(a3D.shape)
cost3D_min=cost3D.min()
c3_argmin=cost3D.argmin()
m,n=divmodc3_argmin,30)
a3_perfect=a3D[m][n]
b3_perfect=b3D[m][n]
ax.plot_surface(a3D,b3D,cost3D,rstride=1,cstride=1,cmap=plt.get_cmap('rainbow'),alpha=.85)
ax.scatter(a_list[0],b_list[0],zs=cost_list[0],s=300,c='r')
ax.scatter(a3_perfect,b3_perfect,c3D_min,s=100,c='green')
ax.set_xlabel('a')
ax.set_xlim((-2,7))

ax.set_ylabel('b')
ax.set_ylim((-2,7))
ax.plot(a_list,b_list,zs=cost_list,zdir='z',c='r',lw=3)
plt.show()
```

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_02.png)

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_03.gif)

**LR=1**

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_04.gif)

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_05.png)

```python
# test 2
y_fun = lambda a, b: a * x**3 + b * x**2
tf_y_fun = lambda a, b: a * x**3 + b * x**2
```

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_06.png)

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_07.gif)

## 局部最优,全局最优

```python
# test 3
y_fun = lambda a, b: np.sin(b*np.cos(a*x))
tf_y_fun = lambda a, b: tf.sin(b*tf.cos(a*x))
```

想象 **np.cos(a\*x)** 是有激活的一层神经层, **np.sin(b\*last\_layer)** 是有激活的第二层.

这个方程就是最简单的一种两层神经网络

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_08.png)

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_01.gif)

**INIT\_PARAMS=\[5, 1]**. 下降到一个最靠近他的局部最优.

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_09.gif)

![](https://morvanzhou.github.io/static/results/tensorflow/5_15_10.png)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://baozoulin.gitbook.io/tensorflow/gao-jie-nei-rong/yong-tensorflow-ke-shi-hua-ti-du-xia-jiang.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.